I’m doing fixel-based analysis on a multi-shell data. Interestingly in three of my subjects, the bias field correction of the preprocessed data with -ants option results in extremely bright voxels in the cerebellar regions, which affects the mask generation (the mask is only generated in the cerebellar region). I used -fsl option and this problem resolved, and the masks were generated covering the entire brain.
The Preprocessed Brain:
Bias Corrected with -ants:
Bias corrected with -fsl:
This was a kind of odd to me and I just wanted to report what I saw.
Bias field correction is fundamentally an ill-posed problem. It’s kind of inevitable that in particular scenarios the process will go awry. We now provide access to the basic ANTS N4 parameters within the dwibiascorrect command-line interface, so you can try mucking around with them; it would also be worth checking the mask that is being provided to N4, as if that omits large parts of the brain then the bias field applied there will be the extrapolation of the field that was estimated elsewhere.